Speaker of Seminar on the New Era of AI-Driven Facility Operations: Innovation and Strategic Collaboration

Speech Title and Abstract

A Chemical Filter Remaining Useful Life (RUL) Prediction Platform Integrating Cross-Concentration Data and AI Digital Twins

As semiconductor manufacturing advances to leading-edge nodes, Airborne Molecular Contamination (AMC) exerts an increasingly severe impact on yield rates. Consequently, precisely predicting the lifespan of chemical filters under sub-ppb (parts-per-billion) ultra-low concentrations has become a critical challenge in facility management. Traditional low-concentration testing is exceptionally time-consuming and costly, making it difficult to reflect dynamic, real-world fab variations in real time.This study proposes an AI-driven Digital Twin predictive framework. It integrates massive high-concentration (ppm-level) breakthrough data generated by a 3,600 CMH high-air-volume testing system at the National Taiwan University Zhubei Campus with traceable national-standard sub-ppb precision baselines acquired via a Mini Enclosure micro-environment platform. The model utilizes an AI-collaborative optimized Lumped Parameter Model, employing machine learning algorithms to process the dynamic evolution of non-linear parameters a (adsorption capacity) and b (kinetic rate) across different concentrations, thereby eliminating the biases inherent in traditional power-function fitting.This approach not only restrains prediction variance to within 15%, but also enables data-driven management tailored to specific process environments. By providing real-time Remaining Useful Life (RUL) estimations for filters, this platform strengthens facility operational resilience and optimizes sustainable replacement metrics, achieving the goals of a new era in smart facility management.